Observation process
- 1
Create hypothetical conversations
We write prompts and follow-up questions based on topics people might ask about.
- 2
Run them through ChatGPT
Each conversation runs in a signed-out session with no account, memory, or previous chat history.
- 3
Add observed ads to the library
If an ad appears, we add the ad and the conversation around it to this library.
How to read the library
What you can take away
- The ad appeared in the recorded conversation.
- The prompt and surrounding chat show the context in which it appeared.
- “First observed here” and “Last observed here” refer only to this library.
- Multiple observations mean we recorded the ad more than once in our sample.
What you cannot take away
- This is not a representative sample of every ChatGPT conversation or ad.
- An ad missing from the library is not proof that it is not running.
- Observation counts are not spend, impressions, reach, targeting, CTR, performance, or market share.
- Viewer geography is unknown. Infrastructure regions are not viewer locations.
